Brown, Kevin S. J., authorFassnacht, Steven, advisorHam, Jay, committee memberMcGrath, Dan, committee memberRoss, Matt, committee member2019-09-102019-09-102019https://hdl.handle.net/10217/197390Seasonal snow is a significant contributor to the water supply of nearly 2 billion people in semi-arid regions around the world. Quantification of this resource is critical to planning sustainable water and food supplies in these regions. While Snow Water Equivalent (SWE) is the most common parameter used to estimate snow water storage, snow depth has often been used as a proxy since it is much simpler to measure and can be converted to SWE if density can be estimated. Depth of snow varies greatly at the regional, watershed, and plot scales and better quantification of this variability can improve water storage estimates. Installation and maintenance of new snow measurement sites is typically expensive and time consuming, so a technology that could produce high temporal resolution snow depth data for a low cost would be useful. Manual reading of snow depth from graduated staffs driven into the ground has been used by the Natural Resources Conservation Service (NRCS) for operational and research purposes. The amount of data available from this method has traditionally been limited by the time-consuming step of manually reading snow depths in images. The central objective of this research was to automate this process in order to reduce the time requirement and allow this technology to be deployed more widely. Five sites were established with time lapse cameras and a set of snow depth staffs around the state of Colorado. Several image recognition methods were considered, and the Aggregate Channel Features technique was used to detect snow depths based on images of the depth staffs. At the most successful sites, absolute error was close to 20 cm, while at less successful sites consistent errors as high as 100 cm made the data unusable. The variety of site configurations examined allowed factors which increased error such as forested backgrounds, close staff placement, and poor camera mounting, to be identified. Additional studies could take advantage of new, cloud-based image recognition technologies in order to allow anyone with a camera and an internet connection to measure snow depth automatically from pictures taken at specific locations.born digitalmasters thesesengCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.measurementtime lapsesnowcomputer visionText